{ "current_gpu_type": "NVIDIA A100-PCIE-40GB", "current_gpu_total_memory": 40339.3125, "perplexity": 18.01548194885254, "token_generation_latency_sync": 59.07657356262207, "token_generation_latency_async": 58.63503850996494, "token_generation_throughput_sync": 0.016927183478912918, "token_generation_throughput_async": 0.017054648984839525, "token_generation_CO2_emissions": null, "token_generation_energy_consumption": null, "inference_latency_sync": 190.97456817626954, "inference_latency_async": 105.70061206817627, "inference_throughput_sync": 0.005236299312257117, "inference_throughput_async": 0.009460683154370061, "inference_CO2_emissions": null, "inference_energy_consumption": null }